search

jmiller656 / EDSR-Tensorflow

Tensorflow implementation of Enhanced Deep Residual Networks for Single Image Super-Resolution
MIT License
330 stars 107 forks source link

after 1 epoch, you don't shuffle? #46

Open Wallace-he opened 5 years ago

kong1060553768 commented 5 years ago

I also see this problem, and'train_set'seems to remain unchanged.

kong1060553768 commented 5 years ago

@Wallace-he Have you solved this problem?

jmiller656 commented 5 years ago

This is correct. I'm actually working on migrating this over to keras (for multiple reasons), this will be one of the problems solved by that. Could use some help figuring out a problem with model saving in keras tho

kong1060553768 commented 5 years ago

Sorry, this is my mistake! Now I want to output the PSNR value in each test. How can I achieve this?

jmiller656 commented 5 years ago

You can watch this using tensorboard, that's how I recommend doing it.

If you want to print PSNR, you'll have to edit the train loop to evaluate PSNR from the graph and print that

kong1060553768 commented 5 years ago

Thank you. I'm sorry to bother you many times. But with your checkpoints and your own training results, the PSNR value of reconstructed images on set5 and test set and individual other pictures is even lower than that of traditional methods. Why?